How to be as 'smart' as Auto-Regressive LLMs:
- memorize lots of problem statements together with recipes on how to solve them.
- to solve a new problem, retrieve the recipe whose problem statement superficially matches the new problem.
- apply the recipe blindly and declare

Post a picture YOU took. Just a pic. No description.

Sylvia Serfaty accepts the 2024 Maryam Mirzakhani Prize from National Academy of Sciences—'I would like to see this as rewarding the pursuit of a personal scientific vision slightly off the beaten path.'

Professor Serfaty's remarks being around 57:07...

EgoPet: Egomotion and Interaction Data from an Animal's Perspective

First purrson view videos.

Problems are solved in the real world after they enter reality, not in an imaginary vacuum. You plan and plan but no plan survives contact with reality. After you release your software or your rocket blows up or your car crashes, you learn and adjust. There is no other way.

I fully agree with Jeremy here. Trying to limit open-source models is very much the wrong way to go, and this California bill should head straight to the circular archive.

There's a new bill, SB-1047 'Safe and Secure Innovation for Frontier Artificial Intelligence Models Act'.

I think it could do a great deal of harm to startups, American innovation, open source, and safety. So I've written a response to the authors: 🧵
answer.ai/posts/2024-04-…

NYU has been named a Nature Index 2023 Rising Star and is ranked 8th in North America for our growth in research published in the world’s leading science journals.

Here are some examples of research work with impact: nyu.edu/about/news-pub…

Important point

My tirade against AI doomers in Davos.
youtu.be/YdaRd_vitLw?si…

martin_casado I think Meta is doing practically everything necessary at this point. Yann LeCun and Mark Zuckerberg are (1) commoditizing everything below the state-of-the-art and (2) they showing the world the safety risks are exaggerated to avoid regulatory capture.

People decrying risk will be proven

As long as AI systems are trained to reproduce human-generated data (e.g. text) and have no search/planning/reasoning capability, performance will saturate below or around human level.

Furthermore, the amount of trials needed to reach that level will be far larger than the

FAIR researchers (AI at Meta) presented SegmentAnything and our robotics work at the White House correspondents’ weekend.

Llama3 + Sim2Real skills (trained with AI Habitat) = a robot assistant

HELM Lite v1.2.0 is out!
Datasets: NarrativeQA, NaturalQA, OpenbookQA, MMLU, MATH, GSM8K, LegalBench, MedQA, WMT14
Results (we still need to add Claude 3, which requires more prompt finagling):
crfm.stanford.edu/helm/lite/v1.2…

It's been a week since LLaMA 3 dropped.

In that time, we've:
- extended context from 8K -> 128K
- trained multiple ridiculously performant fine-tunes
- got inference working at 800+ tokens/second

If Meta keeps releasing OSS models, closed providers won't be able to compete.

.Walkabout Mini Golf™ VR Launches All-New Venice Course, Out Today meta.com/blog/quest/wal…

Knowledge and culture gets lost easily.
Preservation is essential.
Some are trying to sue the Internet Archive out of existence.
Are they modern day barbarians Brewster Kahle ?

It's been exactly one week since we released Meta Llama 3, in that time the models have been downloaded over 1.2M times, we've seen 600+ derivative models on @HuggingFace and much more.

More on the exciting impact we're already seeing with Llama 3 ➡️ go.fb.me/xsqzz8